opencv+hog+svm行人检测自己训练

ID:352148 · 发表于 2018-6-14 21:28

PedestrianDetection-master
1.数据集的预处理\
2.分类器的训练\
3.检测算法的测试\
4.Kalman跟踪\
README.md

全部资料51hei下载地址：

PedestrianDetection-master.zip (76.17 KB, 下载次数: 40)

源码预览如下:

#include <opencv2/core/core.hpp>
#include <opencv2/objdetect/objdetect.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/video/tracking.hpp>
#include <opencv2/video/video.hpp>
#include <stdio.h>
#include <iostream>
#include <fstream>
#include <vector>
#include <algorithm>
using namespace std;
using namespace cv;
//常量定义
char IMG_PATH_TEXT[]="E:/INRIAPerson/color/img_path.txt"; //图像路径
//char IMG_PATH_TEXT[]="E:/研究生课件/机器视觉/Project/Data/pictures"; //图像路径
//const int START_FRAME=243; //开始跟踪的帧243 || 413
//const int END_FRAME=290; //结束跟踪的帧290 || 480
const int WINHEIGHT=180; //图像高度
const int WINWIDTH=320; //图像宽度
const int ROIWIDTH=160; //ROI高度
const int ROIHEIGHT=160; //ROI宽度
const Scalar green(0,255,0);
const Scalar red(0,0,255);
const Scalar blue(255,0,0);
const string DETECT_WINNAME="Detect Window";
const string TRACKER_WINNAME="Track Window";
Mat img(WINHEIGHT,WINWIDTH,CV_8UC3); //the Mat storage the image
Mat blackboard(WINHEIGHT,WINWIDTH,CV_8UC3);
const cv::Rect WHOLEIMG(0,0,WINWIDTH-2,WINHEIGHT-2);
const int STATE_NUM=4;
const int MEASURE_NUM=2;
//使用HOG在img图像中检测行人位置
//返回的矩阵vector为矩阵位置
//返回检测时间
void DetectPedestrian(vector<cv::Rect>& found_filtered,const HOGDescriptor& hog,
double* time=NULL)
{
found_filtered.clear ();
vector<cv::Rect> found;
//hog detector get the cv::Rect with pedestrian
if(time!=NULL)
*time=(double)getTickCount();
hog.detectMultiScale(img, found, 0, Size(8,8), Size(32,32), 1.05, 2);
if(time!=NULL)
{
*time=(double)getTickCount()-*time;
*time /= getTickFrequency();
}
//if the cv::Rect overlap
size_t i, j;
for( i = 0; i < found.size(); i++ )
{
cv::Rect r = found[i];
for( j = 0; j < found.size(); j++ )
if( j != i && (r & found[j]) == r)
break;
if( j == found.size() )
found_filtered.push_back(r);
}
for( i = 0; i < found_filtered.size(); i++ )
{
cv::Rect r = found_filtered[i];
// the HOG detector returns slightly larger rectangles than the real objects.
// so we slightly shrink the rectangles to get a nicer output.
r.x += cvRound(r.width*0.1);
r.width = cvRound(r.width*0.8);
r.y += cvRound(r.height*0.07);
r.height = cvRound(r.height*0.8);
//found_filtered[i]=r;
cv::rectangle (img,r,green,3);
}
std::cout<<" Time : "<<*time<<std::endl;
}
//重载函数，在img的一个子区域中检测行人位置
//该子区域由一个矩形确定
//
//返回的矩阵vector为矩阵位置
//返回检测时间
void DetectPedestrian(vector<cv::Rect>& found_filtered,const HOGDescriptor& hog,
const cv::Rect& roi,double* time=NULL)
{
//CV_Assert(WHOLEIMG.contains(roi.br()) && WHOLEIMG.contains(roi.tl()));
found_filtered.clear ();
vector<cv::Rect> found;
//hog detector get the cv::Rect with pedestrian
if(time!=NULL)
*time=(double)getTickCount();
hog.detectMultiScale(img(roi), found, 0, Size(8,8), Size(32,32), 1.05, 2);
if(time!=NULL)
{
*time=(double)getTickCount()-*time;
*time /= getTickFrequency();
}
Point pt=roi.tl();
//if the cv::Rect overlap
size_t i, j;
for( i = 0; i < found.size(); i++ )
{
found[i]+=pt;
cv::Rect r = found[i];
for( j = 0; j < found.size(); j++ )
if( j != i && (r & found[j]) == r)
break;
if( j == found.size() )
found_filtered.push_back(r);
}
for( i = 0; i < found_filtered.size(); i++ )
{
cv::Rect r = found_filtered[i];
// the HOG detector returns slightly larger rectangles than the real objects.
// so we slightly shrink the rectangles to get a nicer output.
r.x += cvRound(r.width*0.1);
r.width = cvRound(r.width*0.8);
r.y += cvRound(r.height*0.07);
r.height = cvRound(r.height*0.8);
//found_filtered[i]=r;
cv::rectangle (img,r,green,3);
}
std::cout<<" Time : "<<*time<<std::endl;
}
void InitialKalmanFilter(KalmanFilter& kf,double x,double y,
double delta_x,double delta_y)
{
kf.transitionMatrix=(Mat_<float>(STATE_NUM, STATE_NUM) <<
1,0,1,0,
0,1,0,1,
0,0,1,0,
0,0,0,1 );
kf.statePost=(Mat_<float>(STATE_NUM,1)<<
x,
y,
delta_x,
delta_y );
kf.statePre=(Mat_<float>(STATE_NUM,1)<<
x,
y,
delta_x,
delta_y );
//setIdentity: 缩放的单位对角矩阵;
// !< measurement matrix (H) 观测模型
setIdentity(kf.measurementMatrix);
// !< process noise covariance matrix (Q)
// wk 是过程噪声，并假定其符合均值为零，协方差矩阵为Qk(Q)的多元正态分布;
setIdentity(kf.processNoiseCov,Scalar::all(25));
// !< measurement noise covariance matrix (R)
//vk 是观测噪声，其均值为零，协方差矩阵为Rk,且服从正态分布;
setIdentity(kf.measurementNoiseCov,Scalar::all(25));
// !< priori error estimate covariance matrix (P'(k)): P'(k)=A*P(k-1)*At + Q)
// A代表F: transitionMatrix
//预测估计协方差矩阵;
setIdentity(kf.errorCovPost,Scalar::all(25));
}
//获得矩形的质心坐标
Point2f GetCentroid(const cv::Rect& r)
{
Point tl=r.tl();
Point br=r.br();
return Point2f((float)((tl.x+br.x)/2.0),(float)((tl.y+br.y)/2.0));
}
string DescripRect(const cv::Rect r)
{
char buf[128];
sprintf(buf,"This rectangle width: %d , height %d , tl: (%d,%d) ,br: (%d,%d)",
r.width,r.height,r.tl ().x,r.tl ().y,r.br ().x,r.br ().y);
return string(buf);
}
cv::Rect GetROI(const Point2f& centroid)
{
Point tl((int)(centroid.x-ROIWIDTH/2),(int)(centroid.y-ROIHEIGHT/2));
return (cv::Rect(tl.x,tl.y,ROIWIDTH,ROIHEIGHT) & WHOLEIMG);
}
int main()
{
double time;
ofstream fout("data_recorder.txt");
int count=0; //frame count
string img_path; // img path
ifstream fin(IMG_PATH_TEXT); //the text storage the image path
vector<cv::Rect> pedestrian_location;
HOGDescriptor hog; //HoG detector
hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());
cv::Rect preFrame;
cv::Rect curFrame;
int count_tmp=-1;
bool start_track=false; //是否开始跟踪？
int loss_frame=0;
KalmanFilter kf(STATE_NUM,MEASURE_NUM);
vector<Point2f> pedestrian_centroid; //存储质心变化
vector<Point2f> pedestrian_centroid_pre; //预测的质心变化
vector<Point2f> state_post;
//while(getline (fin,img_path) && count<=END_FRAME)
const string video_name = "E:/研究生课件/机器视觉/Project/Data/VID_20141228_151213.mp4";
VideoCapture video(video_name);
Mat frame;
VideoWriter videowriter;
videowriter.open("Test_LX.avi",CV_FOURCC('M','J','P','G'),
10,Size(WINWIDTH,WINHEIGHT),true);
VideoWriter boardwriter;
boardwriter.open("Test_LX_board.avi",CV_FOURCC('M','J','P','G'),
10,Size(WINWIDTH,WINHEIGHT),true);
while(1)
{
//BEGIN !
//Read the image from disk
//img=imread (img_path,1);
video >> frame;
if(frame.empty ())
break;
cv::resize (frame, img, Size(WINWIDTH,WINHEIGHT));
blackboard.setTo (0);
//绘制行人质心
for (size_t k=0;k<pedestrian_centroid.size ();k++)
{
cv::circle(blackboard ,pedestrian_centroid[k],3,green);
if(k!=0)
cv::line(blackboard ,pedestrian_centroid[k],pedestrian_centroid[k-1],
green,2);
}
for (size_t k=0;k<pedestrian_centroid_pre.size ();k++)
{
cv::circle(blackboard ,pedestrian_centroid_pre[k],3,red);
if(k!=0)
cv::line(blackboard ,pedestrian_centroid_pre[k],
pedestrian_centroid_pre[k-1],red,2);
}
for (size_t k=0;k<state_post.size ();k++)
{
cv::circle(blackboard ,state_post[k],4,blue,-1);
if(k!=0)
cv::line(blackboard ,state_post[k],state_post[k-1],blue,2);
}
char buf[128];
sprintf(buf,"frame: %3d",count);
/*
putText (blackboard ,string(buf),cv::Point(10,20),
FONT_HERSHEY_SIMPLEX,0.2,Scalar::all(255));
putText (blackboard ,"Red Rectangle: ROI from Kalman",cv::Point(10,40),
FONT_HERSHEY_SIMPLEX,0.2,Scalar::all(255));
putText(blackboard , "Red Circle: Centroid from Kalman",cv::Point(10,60),
FONT_HERSHEY_SIMPLEX,0.2,Scalar::all(255));
putText(blackboard , "Green Rectangle: Pedestrian Rect from HOG",
cv::Point(10,80),
FONT_HERSHEY_SIMPLEX,0.2,Scalar::all(255));
putText(blackboard , "Green Circle: Centroid from HoG",cv::Point(10,100),
FONT_HERSHEY_SIMPLEX,0.2,Scalar::all(255));
putText(blackboard , "Blue Circle: posteriori state estimate",cv::Point(10,120),
FONT_HERSHEY_SIMPLEX,0.2,Scalar::all(255));
*/
if (!start_track && (count_tmp==-1)) //没有开始跟踪
{
// HoG detection
//这里时候存在行人呢？？
//如果没有行人，就接着读下一帧吧，直到找到
//如果有行人，就目标建立卡尔曼滤波器
DetectPedestrian(pedestrian_location,hog,&time);
if (pedestrian_location.size ()==0) //在该帧没有找到行人
{
cv::imshow (DETECT_WINNAME,img);
waitKey(2);
printf("Frame %d cannot find pedestrian! \n",count);
count++;
continue;
}
else //在这帧找到了行人
{
preFrame=pedestrian_location[0];
//如果有多个矩形框的话，先进行取交集运算
for (size_t j=1;j<pedestrian_location.size ();j++)
{
if (preFrame.area ()<pedestrian_location[j].area ())
{
preFrame=pedestrian_location[j];
}
//preFrame |= pedestrian_location[j];
}
count_tmp = count; ///tmp 赋值
printf("Frame %d find pedestrian! \n",count);
}
cv::imshow (DETECT_WINNAME,img);
}
else //开始了跟踪
{
if (count==(count_tmp+1)) //利用相邻两帧数据建立卡尔曼滤波器
{
cv::Rect roi(WINWIDTH/2,WINHEIGHT/2,WINWIDTH/2,WINHEIGHT/2);
roi=(roi&WHOLEIMG);
std::cout<<roi.tl ().x<<" "<<roi.tl ().y<<endl;
std::cout<<roi.br ().x<<" "<<roi.br ().y<<endl;
DetectPedestrian(pedestrian_location,hog,WHOLEIMG,&time);
if(pedestrian_location.size ()==0) //没有找到
{
start_track = 0;
count_tmp = -1;
continue;
}
curFrame = pedestrian_location[0]; //获得了t+1帧的位置信息
Point2f pt1=GetCentroid (preFrame);
Point2f pt2=GetCentroid (curFrame);
// 1. 初始化
InitialKalmanFilter (kf,pt2.x,pt2.y,pt2.x-pt1.x,pt2.y-pt1.y);
state_post.push_back (pt2);
pedestrian_centroid.push_back (pt2);
pedestrian_centroid_pre.push_back (pt2);
cout<<"Start!"<<endl;
fout<<"Kalman Initial Complete!"<<endl;
cv::imshow (DETECT_WINNAME,img);
}
else
{
// 2.预测
Mat predict=kf.predict ();
Point2f predictPt(predict.at<float>(0),predict.at<float>(1));
fout<<"This is the "<<count<<" frame, and the predict point is ( "<<
predictPt.x<<" , "<<predictPt.y<<" )"<<endl;
pedestrian_centroid_pre.push_back (predictPt);
//cv::circle(blackboard,predictPt,5,red);
cv::rectangle (blackboard,GetROI (predictPt),red,3);
cv::circle (img,predictPt,5,red,5);
// 3.更新Update
// 1) 使用HoG检测
cv::Rect curroi=GetROI (predictPt);
//cv::Rect curroi=WHOLEIMG;
//fout<<"This is the "<<count<<" frame, and "<<DescripRect(curroi);
DetectPedestrian (pedestrian_location,hog,curroi,&time);
// 2)根据位置更新
if (pedestrian_location.size ()!=0)
{
loss_frame=0;
int len=pedestrian_location.size ();
double *dis=new double[len];
int _min=0;
for (int j=0;j<len;j++)
{
Rect r=pedestrian_location[j];
Point cen=GetCentroid (r);
double distan=(cen.x-predictPt.x)*(cen.x-predictPt.x)+
(cen.y-predictPt.y)*(cen.y-predictPt.y);
dis[j]=distan;
if (distan<dis[_min])
_min=j;
}
delete[] dis; //释放内存
cv::Rect detected=pedestrian_location[_min];
#if 0
for (size_t j=1;j<pedestrian_location.size ();j++)
{
if (detected.area ()<pedestrian_location[j].area ())
{
detected=pedestrian_location[j];
}
//detected |= pedestrian_location[j];
}
#endif
Point2f curCentroid=GetCentroid (detected);
pedestrian_centroid.push_back (curCentroid);
cv::circle (blackboard,curCentroid,5,green);
cv::rectangle (blackboard ,detected,green,3);
cv::circle (img,curCentroid,5,green,5);
fout<<"This is the "<<count<<" frame, and the true point is ( "<<
curCentroid.x<<" , "<<curCentroid.y<<" )"<<endl;
Mat measure=*(Mat_<float>(2,1)<<curCentroid.x,curCentroid.y);
kf.correct (measure);
Mat state=kf.statePost;
Point2f stateP=Point2f(state.at<float>(0),state.at<float>(1));
//circle (blackboard ,sta
state_post.push_back (stateP);
}
else
loss_frame++;
if (loss_frame>=3)
{
printf("Lose the obj in frame %d\n.",count);
// mark
start_track = 0;
count_tmp = -1;
continue;
}
cv::imshow (DETECT_WINNAME,img);
}
cv::imshow (TRACKER_WINNAME,blackboard );
}
count++;
boardwriter << blackboard;
videowriter << img;
/*
int c=waitKey();
while (c!=27)
{
c=waitKey ();
}
*/
}
std::cout<<"-----------------------------"<<endl;
std::cout<<"Complete!"<<endl;
fout.close ();
cv::waitKey();
return 0;
}

复制代码

ID:361324 · 发表于 2018-6-28 22:13

你好，请问你是用什么平台实现的？
是用得opencv哪个版本？
特别感谢，望你回复，希望您能上这个论坛并回答这个问题，多谢你。

ID:361324 · 发表于 2018-6-28 22:27

ID:361324 · 发表于 2018-6-28 22:28

请问帖主是采用什么平台运行整个程序的，用的什么版本的opencv

ID:361324 · 发表于 2018-6-28 22:32

你好，请问帖主是在什么平台下运行的程序，用的opencv是什么版本的？

帐号		自动登录	找回密码
密码			立即注册